Hi,
I have a 4790K clocked at 4.8Ghz and 1.2V as 24/7 OC and noticed that when i leave it idle overnight it goes up to 5Ghz all by itself while keeping the voltage, doesn't do this when stress testing.
Never seen this happening before and not confortable with it while having a silent fan profile, readings from HWMonitor, temps are ok so it's probably just spikes.
The mobo is an Asus Gryphon Z87, multiplier and volts are set to manual, turbo boost off and it's about the only things i've touched.
What am i missing?
What are you using to monitor with? I recommend HWiNFO64 (just be sure to police it for bogus sensors and use the hide function).
Is that Z87 flashed to the most recent BIOS? With the latest BIOS, they should pretty much all support the 4790K.
Turbo off, multiplier locked, EIST on, all C-States on, set your PLL and power loadlines to the extreme settings. Someone else can add to that if I missed anything. Do all of this in the BIOS, not via some software program at runtime.
Set your supervisor password in BIOS, enable passworded BIOS. Windows-L every time you walk away from the keyboard. Is someone messing with you? (I used to fire people for not locking their sessions).
A note on monitoring software. All of them do what I am about to describe to one degree or another. Sometimes, even the good sensors in a monitoring program can give an occasional false reading. For instance just today, my minimum VRM input power was -140W in HWiNFO64. My theory on why this sometimes happens with otherwise known good sensors, is that they get read in the middle of an update, and thus the program simply shows invalid data. If this happens in a min/max/avg column, it sticks until the program is reset or restarted.
OCCT for me is practically worthless for measurement. All of the PSU voltages shown are invalid. This varies from motherboard to motherboard.
Oh, and just to be honest, using my personal rules, all cores have to be enabled, with HT, and XMP if you have XMP RAM. What's the point of an OC if you have to hobble the rest of the system? Some gamers can get away with disabling the XMP, but it will impact non-gaming usage, such as a test I did with my RAM at 1333 @ 4.8 GHz, and the old NASA/Cray PITEST systems test (with intel-equiv cray directives, al vectorizing lines uncommented, all scalar lines commented, all vectorizing tunables set to cray X/MP defaults, MX=23, and AVX2 optimized by the intel FORTRAN compiler), took two minutes longer to run at 4.8 GHz / DDR3-1333 than it did at 4.6 GHz / DDR3-2400.
And just in case... What did you use to validate and stress test that 4.8 GHz @ 1.200Vcore? This could be another cause of an invalid reading... At that voltage, I don't see any problem with you using Prime95 Balanced Mode, as long as you have adequate cooling. I don't even get into the 70's at 4.5 GHz @ 1.200Vcore (unconditionally stable) using the Small FFT test. If it doesn't BSOD right away, leave it running no less than two and a half hours. At various predictable times into the test, P95 will switch up what it is doing, and at those times are the most common BSOD points, and it's like clockwork. If you can survive 2.5 hours, you are PROBABLY (but not certainly) pretty damned stable. Only further testing, and extended periods of loaded active use will prove it unconditionally stable. Use the current version of Prime95 for this purpose. If your temps hit 90C, abort the test, and if you have adequate cooling in place already, at that point consider delidding that. If it REALLY is stable at 4.8 GHz @ 1.200Vcore, you have one of the sweetest i7-4790's in existence. If, as I suspect, it hasn't been properly validated, you could be getting random errors in a lot of places, including ones that are not obvious like this, and that includes potentially data being written to disk. Remember, just because you can boot into winblowz, and do some lighter tasks, or even low/moderate load stress tests (OCCT for instance), does not mean you are stable.
As an example, after seven weeks actually today, I only today finished verifying my 24/7's:
--------------------------------------------------------------------------
24/7 safe, unconditionally stable
--------------------------------------------------------------------------
4.4 GHz @ 1.181Vcore / 4.2 GHz @ 1.130Vring / 1.800Vrin
4.5 GHz @ 1.200Vcore / 4.2 GHz @ 1.130Vring / 1.800Vrin
4.6 GHz @ 1.245Vcore / 4.4 GHz @ 1.236Vring / 1.900Vrin
4.7 GHz @ 1.295Vcore / 4.4 GHz @ 1.236Vring / 1.900Vrin
I'm actually on the good side of the bell curve, slightly, by no means outstanding, nice, but not outstanding. I can run P95 small, large, or balanced until the cows come home at those settings, and still stay stable in normal use. Every chip is different, and there are good chips, and then there are one of a kind chips. A 4.8 GHz @ 1.200V at face value is a one of a kind (or damned close), and I hope you can excuse me for doubts on methodology.
If you hadn't stress tested that, that's no big deal. We all have to learn sometime. I would suggest finding the lowest stable voltages for 4.4 GHz and working your way up. As you do so, you will notice things like voltage gaps, and other quirks of YOUR chip, that can tell you with a little practice what is going to be stable and what is not when moving to the next step. Also, it depends on who you are talking to, and do chat with others and learn. Others will tell you that you don't need to run P95 or IBT. I personally disagree with them, because the only way to test is to do a proper full systems test, and that does mean loading every unit in the CPU, including not just AVX, but also AVX2. Only then can you be assured that it will be stable at a setting for any program you run now or in the future. Also, for 24/7, the consensus so far is that with Devil's Canyon, 1.350Vcore is a good working maximum, but, I personally prefer to keep it below 1.300V. Remember, the lower, the better. Keep the higher ones as profiles, for use when the edge is needed, but, the lower the voltage, the longer the chip will last. I don't think you'll find much disagreement on that, no matter if you are talking to the 1.300V crowd or the 1.350V crowd.
That 5.0 GHz in my sig was actually stable enough at 1.425V to actually post some good scores at HWBOT in the previous rookie competition. I do not keep it there. And even THAT was unstable for any serious crunching. It ran their tests just fine though. Anything heavier, and up comes the BSOD 0x101.
Also, there are a small handful of OC forums that have people that actually know what they are doing. I am just starting to get a bit active here, and there are plenty of people here that know their stuff. Out of all the others, I also recommend overclock.net. What i am saying is don't limit your sources, it's a big world. And to get a good grip on what is normal, research, research, research, it can give you good starting points, and you work things to your chip from there. Here's a good one for you, set Vrin to 1.900V, set Vcore to 1.300V, set core multiplier to 4.4 across all four, run P95 for a few hours. If it works, bump the multiplier up by one, lather, rinse, repeat... At the point you get a BSOD (probably a 0x124 or 0x101), at that point, you know that all up to that point is probably stable at or below 1.300Vcore. Now, take each that didn't BSOD, and work the Vcore down. Start by 25mV drops. When it starts giving BSODs, bump it up by ten. Then work it by fives. Do this for each. After this, you probably won't have to use P95 again until you mess with the uncore/cache/ring bus (use small FFT for that), and consensus has it that 1.250V should be a rule of thumb maximum for the ring bus, and always keep Vring BELOW Vcore.
Oh, and P95 is not the end-all. It is a good indicator that you are close to a true unconditionally stable voltage though. Use every tool you can to verify, including the most important one of all actual: use of your computer. Here's my recommended list:
Intel Processor Diagnostic Tool
Intel Extreme Tuning Utility
AIDA64 Extreme
RealBench
Intel Burn Test
Prime95
OCCT
Solidworks systems test
CineBench V15
X.264 Benchmark
and for tools:
CPU-Z (do not trust it's Vcore, it just reads the BIOS setting, not the realtime value)
RealTemp
HWiNFO64
There are others to add to that, but those will get you stable at any OC. Use them ALL. Some will tell you that one or another may not be necessary, but the goal is UNCONDITIONAL STABILITY, and you can only do that by throwing anything you can think of at it, no matter what the gamer kiddie says.
Just make sure you have adequate cooling. I recommend an AIO if you can, and when doing OC testing, until you have it down, set the fans at max. If you still have abnormally high temperatures after getting adequate cooling, and if you have a good chip, there is a cure, but it voids the warranty, and that's to delid it, and use a liquid metal TIM on the die (protecting the substrate board compnents with clear nail polish), and use a good TIM on the top of the IHS. Instant 10-30C drop in temps, with the median being right about where mine was at 14-16C.
